Linear-circular regression

Share:

Description

Fits a Von Mises kernel distribution describing a linear variable as a function of a circular predictor, and boostraps the null distribution in order to evaluate significance of radial variation in the linear variable.

Usage

1
fitlincirc(circdat, lindat, pCI = 0.95, reps = 1000, res = 512)

Arguments

circdat

Numeric vector of radian data matched with lindat.

lindat

Numeric vector of linear data matched with circdat.

pCI

Single numeric value between 0 and 1 defining proportional confidence interval to return.

reps

Integer number of bootstrap repetitions to perform.

res

Resolution of fitted distribution and null confidence interval - specifically a single integer number of points on the circular scale at which to record distributions.

Details

Deviation of lindat from the null expecation is assessed either visually by the degree to which the fitted distribution departs from the null confidence interval (use generic plot function), or quantitatively by column p of slot fit in the resulting lincircmod-class object.

Value

An object of type lincircmod-class

References

Xu, H., Nichols, K. & Schoenberg, F.P. (2011) Directional kernel regression for wind and fire data. Forest Science, 57, 343-352.

Examples

1
2
3
4
5
6
7
8
9
#Example with reps limited to increase speed
data(BCIspeed)
i <- BCIspeed$species=="ocelot"
sp <- log(BCIspeed$speed[i])
tm <- BCIspeed$time[i]*2*pi
mod <- fitlincirc(tm, sp, reps=50)
plot(mod, CircScale=24, xaxp=c(0,24,4),
     xlab="Time", ylab="log(speed m/s)")
legend(8,-3, c("Fitted speed", "Null CI"), col=1:2, lty=1:2)